Undergraduate Economic Review
Volume 12 Issue 1 Article 4
2015
Cost - Benefit Analysis of the German High Speed Rail Network
Martin Dorciak Lewis & Clark College, [email protected]
Follow this and additional works at: https://digitalcommons.iwu.edu/uer
Part of the Regional Economics Commons
Recommended Citation Dorciak, Martin (2015) "Cost - Benefit Analysis of the German High Speed Rail Network," Undergraduate Economic Review: Vol. 12 : Iss. 1 , Article 4. Available at: https://digitalcommons.iwu.edu/uer/vol12/iss1/4
This Article is protected by copyright and/or related rights. It has been brought to you by Digital Commons @ IWU with permission from the rights-holder(s). You are free to use this material in any way that is permitted by the copyright and related rights legislation that applies to your use. For other uses you need to obtain permission from the rights-holder(s) directly, unless additional rights are indicated by a Creative Commons license in the record and/ or on the work itself. This material has been accepted for inclusion by faculty at Illinois Wesleyan University. For more information, please contact [email protected]. ©Copyright is owned by the author of this document. Cost - Benefit Analysis of the German High Speed Rail Network
Abstract This study undertakes a cost-benefit analysis of the German ailwar y market looking specifically at the effects of high-speed rail development on railway passenger subsidies. Using OLS regression analysis, I estimate a demand curve for the German railway network at the route level; this is combined with cost curve estimates to yield a required subsidy for rail development assuming a natural monopoly market structure. I find that an increase in demand as a result of the introduction of high-speed rail technology causes a 23.9% decrease in required rail subsidies.
Keywords Consumer Surplus, Cost Benefit, Cost ffE ective, Elasticity of Demand, Railways, Regional Transportation, Transportation
This article is available in Undergraduate Economic Review: https://digitalcommons.iwu.edu/uer/vol12/iss1/4 Dorciak: German High Speed Rail
1 Introduction
Medium distance travel offers the widest array of choices for travellers
in terms of modes of transport. Within the past 30 years, the advent of
high-speed rail technology has once again enabled rail to be a player in
medium distance travel of 200 to 500 miles, enabling railways to compete
with the currently more appealing modes of cars and planes by reducing the
costs and travel time associated with rail travel. The current developments
in railway transport have allowed rail to increase its modal share relative to
the other two modes, where modal share is the percentage of passengers that
utilize one mode relative to all available options. The main motivation of
this study is to estimate the impact of high-speed rail (henceforth referred to
as HSR) technology on the demand for rail transport and compare this with
the costs of implementing HSR. Then I comment on the viability of HSR as
an alternative option to air and vehicular travel. I will look specifically at
the case of Germany and comment on the effectiveness of the German HSR
experience in terms of costs and benefits.
The roots of HSR analysis come from transport economics and valuations
of travel time and fare prices and how they affect consumer demand. Given a
choice of modes on a given route, consumers choose the mode that minimizes
the total trip cost, which equals the sum of the fare, the cost of preliminary
transit (transit to a station/airport), wait times, and the cost of the actual
travel time itself. In the specific valuation of transportation demand, the
primary method is that of an aggregate transportation model, where ”the
demand for a given [mode in] the travel market is explained as a function of
Published by Digital Commons @ IWU, 2015 1 Undergraduate Economic Review, Vol. 12 [2015], Iss. 1, Art. 4
variables that describe the product or its consumers” (Small, 1992: 6). This
derivation of transport demand can be focused even further as in Domencich
and Krafts (1970) approach of direct demand modeling, where the dataset to
be analyzed is comprised of pairs of locations forming various routes in a city
or region from which travel statistics such as ridership figures, fare prices,
and regional statistics are regressed to determine an explicit demand curve
for the specified region. I will employ this particular method to estimate
demand, focusing on a route level analysis of the German railway network
while also differentiating between routes that have HSR service and routes
that do not.
On the supply side, looking at costs and market structures further reveal
key motivations for this study. The predominant market structure for
railroad systems is that of a natural monopoly, where large economies of scale
and prohibitively high startup costs limit the number of firms to one. Even
then however, the suppliers of rail service typically require subsidization to
prevent travel costs from being prohibitively high for the consumer. Mohring
(1972) provides a theoretical framework for optimal subsidization levels in
public transport and posits that if the average riders opportunity cost of
travel time is below the marginal cost of operating a transport system,
then to make marginal cost pricing viable transport providers would have to
subsidize fares to a level that equals the opportunity cost of the passengers
time. This can also be intuited in the case of the monopolistic railway
networks. But rather than subsidizing a consumers time, production costs
must be subsidized to make monopoly pricing viable.
This issue is certainly relevant to current political debates, with developments
https://digitalcommons.iwu.edu/uer/vol12/iss1/4 2 Dorciak: German High Speed Rail
proposed in California and Texas funded by an 8 billion dollar allotment
from the 2009 American Recovery and Reinvestment Act (Peterman, 2009).
Currently, several Central and Western European countries have developed
HSR networks with varying degrees of coverage that offer travellers overall
travel times which are shorter than flying or driving at comparable or
lower prices for medium distances of 100 to 500 miles with almost exclusive
provision of services through state owned monopolies who in turn subsidize
construction costs and fares with government money. This is indicative
of the motivations behind my analysis, as it is in the interest of national
governments to pursue HSR development policies; I aim to comment on the
viability of these developments.
The core of my analysis will focus on the derivation of the demand
and cost curves through a regression analysis of German rail statistics,
specifically focusing on observations of city pairs where the quantity of
rail transport demanded is represented by the route ridership and price
is represented by the total cost to passengers on the given route. Additional
variables include a dummy variable indicating the presence of HSR on a given
route to differentiate between base level rail demand and high speed rail
demand, average population density of the departure and arrival destinations,
average GDP per capita of departure and arrival destinations, and travel
times and fare prices of the primary competitor in that of air travel. The
analysis combines this market demand curve (equal to marginal benefit) with
costs incurred by the monopolist firm to determine the socially optimal level
of HSR utilization for the German market and comment on discrepancies
between optimal and actual levels of ridership and commensurate subsidy
Published by Digital Commons @ IWU, 2015 3 Undergraduate Economic Review, Vol. 12 [2015], Iss. 1, Art. 4
levels.
The paper will proceed as follows. Section 2 reviews the literature in the
field looking at general theories of transportation economics as they pertain
to institutionally provided transport and subsidization of said transport,
and specific applications of HSR analyses estimating costs, demand, and
societal benefits. Section 3 presents the economic framework typical to the
state monopolist market present in Germany and equates curves in the ideal
model with their function in my analysis. Section 4 presents the data and
regression analysis to derive the demand curve and the specification of cost
curves and section 5 represents these graphically and determines the optimal
level of utilization and compares this with observed levels.
2 Literature Review
Mohring (1972) develops a model that illustrates the role of subsidization
in public transport systems and then applies his model to travel statistics
for city busses in the Minneapolis St. Paul area. Mohring provides a general
base for analysis of transport subsidization, and additionally provides specification
of several key cost variables. First, Mohring notes how transport pricing
deviates from traditional price theory models, where travellers also provide
an input into the final cost calculations in that of their travel time. This
makes the total cost to consumers of utilizing transport the fare that is
charged in addition to the opportunity cost of the journey for the consumer.
The model presented by Mohring is presented below on the folowing page
in Figure 1.
https://digitalcommons.iwu.edu/uer/vol12/iss1/4 4 Dorciak: German High Speed Rail
Figure 1: Bus Transport Subsidies
The commodity detailed in the graph is bus rides; as such, cost is
represented as dollars per bus ride and quantity as bus rides per week, with
curves representing short run and long run marginal costs, long run average
costs, and average variable costs. Mohring assumes this form without alteration
[describes] bus operations that are subject to increasing returns and can be
applied practically to metropolitan transit systems (Mohring, 592, 1972).
Additionally, a demand curve is implied as intersecting marginal costs at
point C. This model reconciles both the inputs of the consumer in that
of travel time and the producer in that of the bus system and widgets
are transformed into journeys. Mohring assumes a consumer provides time
inputs valued at E for point B on the average variable cost curve, noting the
curves representation of consumer valuation of time. Then, Mohring notes
Published by Digital Commons @ IWU, 2015 5 Undergraduate Economic Review, Vol. 12 [2015], Iss. 1, Art. 4
fare price at the level F, where demand intersects marginal costs. This would
generate what Mohring refers to as quasi-rent for bus services of EBCF, but
ultimately this falls short of the costs of fixed inputs, and requires a subsidy
of FCDG to meet those input prices. I will use a variant of this model,
representing a natural monopoly structure to estimate required subsidies to
supplement fare revenues, something I will elaborate on in section 3.
Empirical studies of HSR focus predominantly on route level and network
(national) level analyses. Couto and Graham (2007) conduct one such
analysis, specifically looking at the impact of HSR service on quality and
speed of travel and the commensurate affect on their derivation of a rail
demand function at national levels. The study first posits a basic demand
relationship where a demand for railroad services is expressed as a function
of fare price, level of income, and other factors such as the presence of
HSR (represented as a dummy), geographical and economic conditions such
as regional incomes and city size of destinations, and prices of alternate
modes of travel (Cuoto and Graham, 114, 2007). To estimate their model,
Cuoto and Graham use a log-linear form 2SLS model to estimate coefficients
that are themselves the respective [demand function] elasticities (Cuoto and
Graham, 120, 2007). They regress y passenger-kilometers per kilometer of
network length, (roughly an indicator of network utilization relative to the
size of the network) on four dummy variables representing the introduction
and use of two different HSR technologies (conventional HSR, and tilting
HSR which runs on existing lines).
Additional variables include fare prices, alternate transport prices, and
national geographical and economic indicators. Cuoto and Graham chose
https://digitalcommons.iwu.edu/uer/vol12/iss1/4 6 Dorciak: German High Speed Rail
the 2SLS model due to the structure of their variable representing railway
demand. The demand variable took the form of passenger kilometers per
kilometer of network length, and the 2SLS form was necessary due to the
endogeneity of passenger kilometers (a measure of total kilometers travelled
by all passengers) in their regression. Their regression results indicate
a price elasticity of demand of -0.22, which is in line with estimates of
price elasticity from similar studies by Fitzroy and Smith (1995,1998) and
McGeehan (1984). They further find that conventional high-speed technology
increases railway demand overall at a national level, which coincides with
Fowkes and Nash (1991) who concluded that regardless of speed increases,
demand for rail travel will raise with the presence of a national HSR network.
In turn this suggests a 9% increase in passenger demand for railway travel
given the existence of a HSR network. The variable for tilting high-speed
technology was not found to be statistically significant.
Ultimately Cuoto and Graham conclude that if railway development is
to be supported, then HSR technology is a viable method to increase rail
demand. The study concludes with a brief statement as to the benefits
of HSR investment in that both tilting and conventional HSR can result
in increased demand with tilting technology requiring a much lower initial
investment (due to utilization of exiting networks) but only increasing demand
due to increased frequency, and conventional technology requiring a much
higher investment but resulting in greater overall demand. This is key to my
study and is a central assumption that I will test for the German market.
Rather, that HSR service inherently increases demand on a given route, as
opposed to a network. The magnitude increase estimated by Cuoto and
Published by Digital Commons @ IWU, 2015 7 Undergraduate Economic Review, Vol. 12 [2015], Iss. 1, Art. 4
Graham will also be useful to compare my conclusions to and to note the
difference in magnitudes between route level and network level analyses.
Behrens and Pels (2010) look specifically at intermodal competition
on the London Paris corridor where they work to determine passenger
choices between HSR and air travel from passenger preferences as observed
surveys and explicit ridership data. Behrens and Pels utilize nested logit and
mixed logit models to explain passenger choice on the London Paris route.
Their study draws from past work utilizing logit regressions to determine
passenger behavior, and draw their key variables from these studies in that
of fares, accessibility (to transport), frequency (of offered journeys on a
route), and reason for travel (leisure or business). Specifically, Behrens and
Pels use similar characteristics in that of travel costs and times for rail
and substitutes, frequency, delay times, and personal statistics; they also
differentiate regressions between leisure travellers and business travellers
under the assumption of different values placed on particular characteristics.
The model specified takes two forms, first a nested form where choice of
mode is nested first by type (air or rail) and then by specific route (arrival
and departure airport pairs and different HSR services). The second form
merely groups all the possible routes into one group and assumes consumers
choose mode and route in one step. Their results concluded that HSR is a
viable competitor for air travel on the specified route. Business travelers in
particular were found to choose HSR due to their frequency of service over air
travel. Conversely, it was found that leisure passengers tended to substitute
into HSR in the face of rising airfare costs, and the relative stability of HSR
fares. The authors additionally note that several of the flights (by various
https://digitalcommons.iwu.edu/uer/vol12/iss1/4 8 Dorciak: German High Speed Rail
airlines) between the cities that they used in the model have since been
discontinued, showing the practical effects of passengers choosing HSR over
air travel on routes where airlines encounter strong competition from rail.
Both this study and Cuoto and Graham provide insight into my own study
regarding how to approach the functional form of my demand regression as
well as detailing key variables of interest.
On the supply side, Levinson, Mathieu, Gillen, and Kanafani (1997)
conduct a theoretical study to estimate the full costs of developing a high
speed rail corridor in the state of California based on past observations and
economic theory. Their primary motivation lies in determining the private
and social costs of providing an intercity HSR network and to determine
what these costs imply when looking at HSR investment as opposed to
air travel and highway infrastructure investment. One of the key issues
with such an analysis as noted by the authors is the differentiation between
costs paid by the users of the transport mode and costs paid by others.
Their taxonomy for full cost to society (FC) includes numerous variables
but chiefly among them are infrastructure costs for construction (IC) and
maintenance (MC) of the rail network, carrier capital and operating costs
incurred by operators of rail service for the purchase of vehicles and other
items (CC), user money costs (fares and fees, FF), user travel time costs
(opportunity cost of passengers, represented by TC), user delay costs (opportunity
cost of delays, DC), and social costs incurred by people exogenous to the
system in that of emissions and noise pollution (SC). The full cost of high-speed
rail development and operation on a given route is thus calculated as a sum
Published by Digital Commons @ IWU, 2015 9 Undergraduate Economic Review, Vol. 12 [2015], Iss. 1, Art. 4
of the aforementioned values and takes the following form.
FC = IC + MC + CC + FF + TC + DC + SC (1)
The authors follow this specification with the caveat that ”each cost is
a function of various parameters and depends on the level of [ridership]”
and devote the remainder of the paper to estimate these costs as functions
of ridership (Levinson et al, 192, 1997). The infrastructure costs are stated
as the costs of building the rail network and all that entails in terms of
structures, rail, power, and earthworks; this is simply divided by the number
of passengers utilizing the system to determine costs per passenger and leads
to a negatively sloped average infrastructure cost as ridership increases.
Carrier costs follow and are divided into two parts, namely carrier operating
and capital costs. Both are determined by multiplying the cost per unit by
the number of units in operation (where unit refers to a train) for operating
expenses such as electricity and worker salaries, and capital expenses in
that of the cost of the train. These costs will only vary with the number
of units in service and assuming the rolling stock of rail companies remains
constant in the short run, the marginal costs of accommodating additional
travellers on a given train is essentially fixed. This study is unique among
others in that the authors account fully for all costs incurred whereas other
studies focus only on some sources of costs and quantify them differently
(such as cost per passenger-kilometer, or passenger journey), Levinson et.
al. instead focus on the full costs incurred by society. The remaining costs
are noted as incurred by the consumer and the public in general. The
https://digitalcommons.iwu.edu/uer/vol12/iss1/4 10 Dorciak: German High Speed Rail
authors posit theories for the remaining factors where noise and pollution
can be interpreted by variations in housing price around HSR lines, but this
section is generally speculative.
Levinson et. al. use these costs to determine the total cost per user
of the system and compare it to the costs of air and vehicular travel.
Additionally the authors continue to tie their estimates back to the proposed
Californian HSR network. Their analysis and cost calculations are based on
data from the European Union and Japan, however the authors acknowledge
the differences and difficulty in comparisons of developments in other countries
and the United States. The authors conclude by noting that it would be
difficult to implement HSR in California without massive subsidization and
even then so with higher subsidization than other countries. This is due
to constraints on federal spending, the deregulated nature of air transport
providing cheaper airfares (which was largely state directed in Europe and
Japan during the advent of HSR), and the sheer geographical distances
between Californian cities. However, the cost curve specifications in this
model can be incorporated into the theory of Mohrings model and applied
to the German case.
3 Economic Theory
The essential theory of my analysis requires several preliminary model
specifications and assumptions. First, it is important to note the prevailing
market structure in HSR before pursuing further specificity. As is the case of
most public transport goods, HSR networks are integrated into state owned
Published by Digital Commons @ IWU, 2015 11 Undergraduate Economic Review, Vol. 12 [2015], Iss. 1, Art. 4
railway companies and take the form of a natural monopoly (Mohring, 1972).
Specifically, the railway industry has very high costs of production with the
good in question being transport, but small and constant marginal costs;
also featured are large economies of scale and a large infrastructure that is
consolidated under the monopolist firm.
3.1 German Railway Market Structure
In the case of Germany, national transport and freight railroad is consolidated
under Deutsche Bahn AG (henceforth referred to as DB), a private joint-stock
company with 100% ownership by the German government (Deutsche Bahn).
Dunn and Perl (1994) provide further information on the formation and
structure of this network; DB stretches back to the Cold War and the
reunification and integration of the two Germanies and how this effected
the formation of a national railway system. This national disunity could
explain the slow pace of the introduction of Germanys high-speed Intercity
Express (ICE) train network. Additionally, the paper provides other details
that are not mentioned in economic analyses that shed light on some added
costs and rationale as to why the German network took the form that it
did. The primary difference between German and other HSR according to
the authors is the German utilization of standard previously existing track
for both passenger and freight transport upgraded to handle HSR trains, a
choice made to avoid prohibitively expensive costs of tunneling new HSR
lines through mountains. This has several economic implications, namely
the cost savings as opposed to constructing a new network, and the slightly
slower travel times on German HSR relative to France and other nations.
https://digitalcommons.iwu.edu/uer/vol12/iss1/4 12 Dorciak: German High Speed Rail
All of this can be combined into a general model for a given railway
route in Germany that I will use in my study, presented below in Figure 2.
This model takes an adaptation of the general form presented by Mohring
Figure 2: Railway Route Model
(1972), where D = MC is the fair market fare price including time valuation,
and MR = MC is the fare price charged. MC in this case is fixed due to
the fixed nature of accommodating an additional passenger on rolling stock
that has already been purchased, this is to say, assuming a fixed number
of trains running a fixed number of routes, the cost of accommodating an
additional passenger on a train is constant. Marginal revenue is derived
Published by Digital Commons @ IWU, 2015 13 Undergraduate Economic Review, Vol. 12 [2015], Iss. 1, Art. 4
from the demand curve by transforming demand into total revenue and then
taking the derivative, from which I can determine the fare price charged on a
given route by looking at the intersection of marginal revenue and marginal
costs. As in a traditional monopoly, this determines the price and from
that I can determine realized producer surplus as indicated above in Figure
2. This however falls short of the AFC curve, which here represents the
infrastructure costs per rider incurred by DB in the construction of rail
lines and modification of existing lines to accommodate high-speed trains.
It is here that Mohrings theory comes in to play with DB being unable to
cover total costs at any price or ridership level thus requiring subsidization to
cover all costs. As follows, the distance between D and AC at the monopolist
price level (the distance between Fare and Total Cost in Figure 2) is equal
to the subsidy needed to cover all costs that are not incurred by riders at
the monopolist price level. Additionally important to mention, due to the
nature of my data this model compares costs and demand at the route level,
where a route is defined as a path of travel between a city pair, rather than
looking at the railway network in Germany as a whole.
This ultimately leads to the key point of interest in my study. A central
assumption to my model is borrowed from Mohring in that of an average
cost curve that is higher than demand for all quantities of passengers, and
thus unlike a standard monopolist firm, railroad providers cannot charge a
fare that will recoup all infrastructure and rolling stock costs and be low
enough to maintain any passenger levels. In regards to HSR, Cuoto and
Graham (2007) concluded that the presence of HSR led to an across the
board increase in demand for rail service, in line with other studies. In the
https://digitalcommons.iwu.edu/uer/vol12/iss1/4 14 Dorciak: German High Speed Rail
model this translates to a shift outward in the demand curve for a network
with HSR service and as such leads to a higher ridership level and a lower
required subsidy. This change in required subsidy level will be the key focus
of my analysis, looking at if the introduction of HSR provides any meaningful
magnitude change in the level. The next two subsections will focus on the
specification of the functional form of the demand and cost curves used in
my model.
3.2 Railway Demand
The central component of this study focuses on estimating a demand
curve at the route level of the German railway network. I will estimate a
linear demand specification for a given German route that takes the form
of a traditional linear demand curve. The components that determine the
curve are total cost (TC) of a rail journey to a consumer and quantity of rail
travel demanded (D), which is represented by the number of rail passengers
on a given route as used by De Rus and Inglada (1997) to estimate railway
demand in Spain. As such, the relationship of interest is how the number
of rail passengers responds to changes in the total cost of a given journey.
The specific functional form of the regression utilized is as follows.
D = β0 + β1TC + β2X2 + ... + βnXn + (2)
This again represents the form assumed above, where railway demand is
dependent on total cost of travel, with other explanatory variables also
included. This regression yields an inverse function explaining the change
Published by Digital Commons @ IWU, 2015 15 Undergraduate Economic Review, Vol. 12 [2015], Iss. 1, Art. 4
in quantity demanded as a function of price. The demand function can be
transformed,
β 1 TC = 0 + D (3) β1 β1
. This transformation allows my demand function to be represented graphically
alongside the remaining curves in my model by standardizing axis variables
and by retaining the relationships of interest yielded in the regression results.
3.3 Railway Costs
Costs in this model are relatively straightforward, and again are represented
as the costs to consumers at the route level. The unique aspect of Germany
and its usefulness in regards to this study is that of the nature of its
railway system. The benefit lies in the fact that Germany uses the same
track network for both normal and HSR applications, with both freight
and passenger traffic also travelling on the same track. This is also a key
determinant of which railway lines in Germany received HSR upgrades;
while major population centers will be linked with high-speed service, the
unification of freight and passenger traffic gives incentive to bring higher
speed service to industrial centers that may not have much consumer demand
for the service (Albalate and Germa, 2012). This would seem to further
support the choice of Germany as an ideal location for this study, where
passenger travel would seem to take a secondary role when it came to
developing HSR lines to that of freight transport, thus leading to the conclusion
that there may be some randomness in the distribution of the HSR treatment
https://digitalcommons.iwu.edu/uer/vol12/iss1/4 16 Dorciak: German High Speed Rail
in my model (notwithstanding highly likely endogeneity issues between freight
and passenger transport between cities).
Ultimately, this suggests however that a route level analysis and comparison
of costs is viable as much of the German railroad network thus has a uniform
average cost per kilometer of track. First, it is necessary to specify the
average costs per passenger incurred by DB on infrastructure costs on a
given route. I am assuming variable costs to be negligible in my model,
as the two major cost groups (infrastructure and rolling stock) are both
fixed for a given route, thus average costs would simply be fixed costs of
infrastructure per route divided by passenger level. This simplification
allows me to assemble a variable for route cost (RC) as the product of
average cost per kilometer and route length. This cost per route must then
be divided by the amortization period (AP) of the route development costs.
This is then divided by passenger level (n) to determine the average fixed
cost per passenger at a given passenger level. This is represented as follows.
RC AC = AP (4) n
Additionally, since passenger level (n) is variable, this equation will take a
downward convex form. This is intuitively sound and can even be equated
to a long run average total cost exhibiting increasing returns to scale, given
my assumption of negligible short run variable costs.
Marginal costs in my model are held fixed. This is rationalized due to the
fixed capacity of the German rail system in that once trains are acquired, the
cost of accommodating an additional passenger on a train is fixed up until
Published by Digital Commons @ IWU, 2015 17 Undergraduate Economic Review, Vol. 12 [2015], Iss. 1, Art. 4
the capacity of the system is reached and more trains must be purchased.
Currently, the German rail system is comprised of refurbished trains from
the 90s, newer low cost regional trains, and 67 high speed trains built by
Siemens AG. Even in this regard, Germany is ideal for this study as rather
than purchasing trains on an as-needed basis, DB purchased the trains in
two major transactions, one in 2000 acquiring 50 Siemens Velaro ICE 3 Class
trains with 30 years of maintenance and a second in 2008 with 17 Siemens
Velaro D Class trains for international use based in Germany (Siemens,
2008). Thus to determine the marginal cost per passenger I take the total
cost (TC) of all rolling stock acquired by Deutsche Bahn and divide it by
the number of years in the period of amortization (AP) of the funds used
to purchase the trains. This is then be divided by the annual maximum
theoretical ridership (total train capacity multiplied by number of trains
multiplied by journeys per day) (TR).
TC MC = AP (5) TR
The value represented here again is the marginal cost of accommodating
an additional passenger on a train to recoup the costs of the train and its
maintenance.
4 Empirical Application
My first application is to estimate the demand curve for German rail
travel as detailed above in my theoretical discussion.
https://digitalcommons.iwu.edu/uer/vol12/iss1/4 18 Dorciak: German High Speed Rail
4.1 Demand Curve Regression
The majority of the data being used in my analysis was acquired from
Eurostat, a Directorate General of the European Union tasked with gathering
and providing data on many societal aspects to institutions of the EU and
other parties. Data collected by Eurostat was most recently updated in a
complete state in 2010. The database contains travel statistics for all cities
in the countries of interest and allows the user to pair a departure city with
an arrival city and shows the total traffic between the two cities on a given
mode of transport in the survey year. From this, I assembled 100 pairs
of arrival and departure cities, half with high-speed rail service and half
without. These pairs were determined by looking at a map of the German
rail network and assembling the pairs, additionally a route was only given a
designation of ”High Speed” if the route had the high-speed service in 2010
(Raileurope). I then retrieved passenger data for my city pairs on both rail
and air travel; I also retrieved density and GDP per capita data for each
city, all from Eurostat. Fare price data was also collected for each route
for rail transport and air transport. The explicit variables to be used in
my regression and their meanings are presented in Table 1 along with their
sources. Totrailcost and totaircost consist of two components, first is the
base fare charged to passengers for their journey, then added to this is the
opportunity cost which is simply calculated as the time a traveller could be
spent working, found by taking the average GDP per capita of the city pair
and dividing that by an average annual workload (8 hours a day, 5 days a
week). This is then divided by 60 to get the opportunity cost per minute,
Published by Digital Commons @ IWU, 2015 19 Undergraduate Economic Review, Vol. 12 [2015], Iss. 1, Art. 4
Table 1: Regression Variables
Variable Definition railpass (Eurostat: Railway Transport) Total railway passengers between the given city pair totrailcost (Deutsche Bahn) The fare cost and opportunity cost of a rail journey in dollars totaircost (Google) The fare cost and opportunity cost of an air journey in dollars density (Eurostat: Density) The average population density of the given city pair in people/SqKM gdpcap (Eurostat: GDP per capita) The average GDP per capita of the given city pair in Dollars hsr A dummy variable indicating the presence of HSR service on a given route
and multiplied by the number of minutes in the journey for both air and rail
transport.1 Determining railpass merely involved assembling the city pairs
in Eurostat and inputting the values into my own dataset. Density and
gdpcap were also straightforward and represent the average of the respective
metric in the destination and arrival city. Lastly, hsr was determined by
seeing which city pairs were positioned on a stretch of high speed line and
assigning a dummy value of 1 for high speed service and 0 otherwise2.
Summary statistics for the variables of interest are presented in Table
2. The first apparent observation that can be made is in regards to costs,
1For rail travel time, the total time in transit was merely considered. For air travel time, I added two hours to the total flight time to account for checking in and traveling to airports typically on the outskirts of cities. 2In the rare case that a route contained sections both on high speed and normal speed track, the dummy was assigned to the simple majority.
https://digitalcommons.iwu.edu/uer/vol12/iss1/4 20 Dorciak: German High Speed Rail
Table 2: Summary statistics
Variable Mean Std. Dev. N
railpass 417255 681150 100
totrailcost 281.03 83 100
totaircost 392.47 134 100
density 2442 630 100
gdpcap 35963 5157 100
hsr 0.49 0.502 100
with air costs at a higher average level than rail costs. Air costs also vary
to a greater extent than rail costs with a higher standard deviation at $134
versus $83 for rail. This can be accounted for by looking at some of the
lower traveled city pairs, which can still sustain a rail link due to the low
marginal costs of rail operation given existing track between the two cities.
Realistically, Deutsche Bahn can raise prices on such routes to still keep
them affordable, but to offset potential lower demand; air companies cannot
do the same and must charge a higher price to recoup costs of flying a low
demand route, should the plane not be full. Density and GDP per capita
statistics also fall in line with expectations, with GDP per capita being fairly
homogenous and density being decidedly less so3.
I will utilize an OLS multiple linear regression as specified in equation (2)
of my theoretical framework. Including the above variables, the regression
3This is as a result of looking at the standard deviations and means with standard deviation being 14.3% of the mean for gdpcap and 25.8% of the mean for density.
Published by Digital Commons @ IWU, 2015 21 Undergraduate Economic Review, Vol. 12 [2015], Iss. 1, Art. 4
will take the following form.
railpass = β0 +β1totrailcost+β2totaircost+β3density+β4gdpcap+β5hsr+
(6)
The initial regression indicated a heteroskedastic relationship after applying
the Breusch-Pagan Test, thus the following regression summary presented
in Table 3 was corrected for heteroskedasticity and is presented with the
robust standard errors.
Table 3: Estimation results : Route Demand
Variable Coefficient (Std. Err.) Prob. T
totrailcost -3830 (585) 0.000
totaircost 2038 (372) 0.000
density 361 (82) 0.000
gdpcap 12 (10) 0.232
hsr 221484 (103346) 0.035
Intercept -748015 (445370) 0.096
N 100
R2 0.53
F (5,94) 21.212
The regression results indicate a relationship in line with the expectations
of my model, in that a dollar increase in fare price on a railway route
https://digitalcommons.iwu.edu/uer/vol12/iss1/4 22 Dorciak: German High Speed Rail
will lead to a statistically significant 3830-passenger decrease in annual
ridership. Also of inerest is the coefficient on hsr, which again is in line
with the expectation in that the presence of HSR is estimated to increase
annual ridership by 221,484 passengers significant at 98.2% confidence. The
variables representing total air cost and density can also be rationally explained.
First, the coefficient on totaircost is positive and indicates an increase in
total air cost of 1 dollar causes substitution of 2038 passengers per year into
the competition in that of rail travel, and is significant at 97.2% confidence.
Additionally, density also shows a positive coefficient of 361 additional annual
rail passengers for an additional person/km2 of average density at more
than 99% confidence. This can again be rationalized by the notion that
the larger average density of the route would imply a greater need for
travel among residents. Lastly, the coefficient on gdpcap is not found to
be statistically significant. An observation of a scatter plot showing the
relationship between gdpcap and railpass indicates a relatively uniform level
of rail ridership at all levels of GDP per capita which is intuitively sound
when the historical precedent of heavy railway use among Germans is taken
into account, particularly among the middle class. This is to say that
railway transport is and has been marketed as a mode of transport for
all socioeconomic bacgrounds with class based fares, more akin to air travel
than public transport. Regardless, the variable will still be included to avoid
potential omitted variable bias.
Shortcomings of this portion of the analysis are most evident when
looking at the goodness of fit of the linear model estimated. The most
straightforward way to check for this is through an application of the Ramsey
Published by Digital Commons @ IWU, 2015 23 Undergraduate Economic Review, Vol. 12 [2015], Iss. 1, Art. 4
RESET test. The test yields an F-value of 20.37, which when compared
to an F-statistic of 2.71 at 3 and 91 degrees of freedom indicates that
a non-linear relationship has some explanatory power in determining the
dependent variable. This is further consistent with a scatter plot of the
two main variables of interest, totrailcost and totrailpass as seen below
in Figure 3. A simple observation of the scatter plot shows that there
Figure 3: Totrailcost and Totrailpass Scatter Plot
is indeed a negative relationship between rail costs and passengers, which
seemingly supports the linear estimation given by the regression. However,
the rapidly increasing passenger levels correlated with decreasing costs could
indicate an exponential or logarithmic relationship between the primary
variables of interest. I will however continue with my linear estimation of
the demand function with the justification of it being simpler to integrate
a linear demand function into my model. Additionally, the RESET test
does not preclude a linear relationship but merely indicates that other
https://digitalcommons.iwu.edu/uer/vol12/iss1/4 24 Dorciak: German High Speed Rail
specifications may be superior, and as such I will continue with the linear
model.
Lastly, misspecifications in the model and regression could be caused
by problems of multicollinearity, which should not arise due to the nature
of my data, but will be looked at anyway. Firstly, the variables do not
exhibit strong correlation to each other, and most are again statistically
significant. I also calculated the variable inflation factor for the variables,
and this yielded values ranging from 1.02 to 1.24 for the variables in the
model. These values are interpreted by taking the square root, which then
gives the factor by which the standard error of a variable is inflated compared
to what it would be if there was no correlation between that variable and
others. In the case of my data, these values range from 1.01 to 1.11, and
are far below the accepted threshold for high multicollinearity of VIF = 5.
This combines to allow me to reasonably conclude that multicollinearity of
my variables is not an issue in my regression.
4.1.1 Demand Curve Estimation
I now specify the explicit demand curve according to the method laid
out in Section 3.2 and equation (3).4 Again, the regression I ran above
estimates the effect of total costs on ridership levels, and would place cost
as independent and ridership as dependent if merely placed into a demand
function, essentially showing the inverse demand function. I could opt for
4Again, with the values from the regression being transformed according to β 1 TC = 0 + D (7) β1 β1
Published by Digital Commons @ IWU, 2015 25 Undergraduate Economic Review, Vol. 12 [2015], Iss. 1, Art. 4
regressing all the variables on price instead to bypass this step but the
coefficients would be different and not represent the relationship that I am
ultimately after. Additionally, my model utilizes the other control variables
to determine a more precise estimate for the coefficient on railpass and the
alternate model would not indicate this effect in my desired relationship.
To continue, according to equation (2) my demand function for German
railway routes is,
1 TC = 195 − n (8) 3830
with TC representing the total cost of rail usage, and n representing the
passenger level. This simply indicates that starting at a peak price of $195
(vertical intercept), a one-dollar decrease in price will increase ridership
by 3830, to a horizontal intercept of 746,850 passengers. The inclusion of
the effects of HSR into this will shift this demand curve upward to a new
vertical intercept of 253, while still retaining the same slope and pushing the
horizontal intercept out to 968,990. This comes to indicate that the presence
of HSR on a given route will increase demand for rail travel on that route
by approximately 23%.
This is greater than the estimated increase in demand of 9% referenced
by Cuoto and Graham earlier, but we must remember that this level was
estimated for the impact of HSR on national railway demand rather than
route level demand. This however is not out of line with intuition, as a
national presence of HSR might be small relative to the size of a network.
Thus only a small portion of the population that lives along the routes that
https://digitalcommons.iwu.edu/uer/vol12/iss1/4 26 Dorciak: German High Speed Rail
HSR services and has access to it and their increased demand would be
counteracted by the relatively stable demand on other routes with no HSR
service. Conversely, my higher estimate for route level demand increase
reflects the (reasonably) larger increase in demand of people who have direct
access to the good. This demand function will be revisited later when the
full model is assembled, and the following section will focus on determining
the cost curves in my model.
From this demand curve, I can also simply determine the marginal
revenue curve necessary for the model through basic economic relationships.
Given that total revenue = price x quantity,5 I can take my demand function
and multiply by quantity as follows to determine total revenue, multiply
quantity through the equation, and take the derivative to yield marginal
revenue as follows.
1 1 TR = TC = (195 − n)n or T R = 195n − n2 (9) 3830 3830
δT R 1 MR = = 195 − n (10) δn 1915
This marginal revenue curve maintains the same intercept as the total cost
but with half of the slope and represents the marginal revenue of each
additional passenger on a given route. It will be revisited and incorporated
into the overall model to determine the monopolist price level, and the
required subsidy level.
5With my variables of TC and n equaling price and quantity respectively.
Published by Digital Commons @ IWU, 2015 27 Undergraduate Economic Review, Vol. 12 [2015], Iss. 1, Art. 4
4.2 Cost Curve Estimation
The estimation of cost curves for the model is generally fairly straightforward,
but still require some explanation. With demand estimated, I will first focus
on average fixed costs. In my case, AFC is being determined purely as the
infrastructure construction costs necessary to build a given routes track at
a given passenger level. This can also be equated to overall average costs,
as I noted in Section 3 with my assumption of negligible variable costs. To
reiterate, Germany is an ideal choice on these grounds for this analysis as
all German trains have the ability to run on all types of track, unique to the
German rail network and allowing me to utilize a simple average construction
cost per mile of track to apply to my entire analysis. First, to determine
a total variable for route cost I took the average cost per mile in dollars of
the network (23.1 million per mile) and multiplied this cost by the length
of each route in miles to yield a variable for total route cost (Feigenbaum,
2013). These total route costs are then divided by 10 years, which is the the
amortization period of loans used to fund the construction and the average
is taken, giving an average route cost per year of $54.18 million.
To determine the actual formula for AC I will follow the method specified
in section 3.3 and following equation (4). This simply takes the above value
of average cost per year and divides it by passenger level (n) to determine
the average costs at a given passenger level. The explicit form taken by the
AC curve is as follows.
54180000 AC = (11) n
https://digitalcommons.iwu.edu/uer/vol12/iss1/4 28 Dorciak: German High Speed Rail
This gives a downward sloped and convex AFC curve, which as discussed
previously, can be equated to ATC, given my assumption of negligible variable
costs. The curve itself determines the necessary subsidy level, with the
monopolist passenger quantity being taken up to the AFC curve, which gives
the price level necessary to charge riders to recoup infrastructure costs.
Marginal costs are held constant in my model and are represented as the
cost of accommodating an additional passenger on any given route covering
the costs of rolling stock and maintenance. This fixed level of marginal
cost per route passenger will be determined according to equation (5) that
again takes the total cost of all rolling stock and divides it by the number
of years over which the funds used to purchase the trains are amortized.
Then the cost per year is divided by the maximum theoretical ridership to
determine the marginal cost of accommodating an additional passenger. The
calculation is very straightforward, but values and intuition of results are
still necessary to discuss. Values for the cost of train units are determined
from Siemens AG, the provider of all rolling stock. Converted into dollars,
the approximate cost per train unit purchased by DB is $41,500,000 and
is paid for over 10 years with maintenance included in the purchase price
(Siemens, 2008) Additionally necessary to account for are energy and labor
costs per train which are estimated to be $1,168,000 per train in energy
and $800,000 per train in labor costs which are added to the cost per unit
(Feigenbaum, 2013). This is then multiplied by 67, or the number of recently
(since 1998) purchased train units in operation to yield a total cost of rolling
stock and maintenance of approximately $2,912,356,000 which when divided
by 10 years gives an annual cost of $291,235,600.
Published by Digital Commons @ IWU, 2015 29 Undergraduate Economic Review, Vol. 12 [2015], Iss. 1, Art. 4
The total annual cost must then be divided by maximum theoretical
annual ridership to determine the marginal cost of an additional passenger.
The maximum capacity of one DB train unit is on average 430 passengers,
and in between trains in maintenance time, and rotation of older train stock
on routes, I will assume each new train makes an average of one journey per
day. The average capacity per train of 430 is multiplied by 67 trains and 365
days for a total of 10,515,650 possible passengers in a given year. When the
annual cost is divided by the maximum possible ridership it is determined
that the marginal cost of an additional passenger is $27.70. Intuitively, this
indicates that for any additional passenger on any route, $27.70 of their fare
price will always be required to cover the train units and their operation.
The next section will focus on incorporating this marginal cost curve and
all previous curves into my model and determine the change in subsidy level
resultant from a shift in the demand curve brought about by the presence
of HSR.
5 Cost-Benefit Analysis
5.1 Subsidy Calculation
Now, having derived all the necessary curves, I can assemble the following
graph for my model, representing a given German railway route without
HSR service, showing demand and costs. To note, the graph is not drawn
to scale due to the high passenger levels relative to costs. The main purpose
of this section is to look at the subsidy levels required to make a given
route financially viable by offsetting infrastructure costs and supplementing
https://digitalcommons.iwu.edu/uer/vol12/iss1/4 30 Dorciak: German High Speed Rail
Figure 4: German Railway Route- No HSR
fare revenues. In Figure 4, this is represented as the shaded area marked
subsidy, comprising of the difference between fare revenues (FR) at the
monopolist price and the average cost curve. To begin the calculation, I will
first determine the equilibrium level of passengers n*, on a given route by
setting marginal revenue equal to the marginal cost of 27.7 and solving for
n, which in this case yields a value of 320,380 passengers. This n* is then
inputted into the demand and AC equations to give the equilibrium fare and
infrastructure cost not covered by fare revenue which are $111.35 (Fare*)
and $169.11 (C*), respectively. Next, I will determine fare revenue by taking
Published by Digital Commons @ IWU, 2015 31 Undergraduate Economic Review, Vol. 12 [2015], Iss. 1, Art. 4
the difference of Fare* and marginal costs to give a revenue per passenger
of 83.65, this is then multiplied by the equilibrium ridership level n* to give
a fare revenue of approximately $26.8 million per annum on a given route
without HSR, represented in Figure 4 by the shaded region marked FR. The
required subsidy level is calculated similarly by taking the difference of C*
and Fare* and then multiplying by n* to get a reqiured annual subsidy of
$18.505 million per annum on a route.
The main statistic that this boils down to is that for a given non high
speed route in Germany, an annual subsidy of 34.2% of total costs is required
to supplement fare revenues and break even at the equilibrium price and
ridership level. In hindsight of looking at several other studies of current
required transit subsidies, my estimate would seem to understate the actual
level of subsidization present in much of Europe and America. For example,
rail projects in the UK and France typically subsidize total costs by two
thirds to arrive at the final passenger fare (European Environment Agency,
2007). In America the levels are even higher, with an average of 70-80% of
costs being subsidized on commuter rail projects (Garrett, 2004). This can
lead to two conclusions, first that my figures and model are miss-specified,
which may account for some of the variation due to previously mentioned
simplifications I make, such as assuming demand to be linear and variable
costs to be negligible. Second, the nature of these cost-benefit analyses
in general may result in an understatement of the required subsidy levels
(Utsunomiya and Hodota, 2011). This is to say that all cost-benefit valuations
of existing systems undertaken to make future policy suggestions are chronically
unable to account for the full costs of rail development in terms of cost
https://digitalcommons.iwu.edu/uer/vol12/iss1/4 32 Dorciak: German High Speed Rail
overruns, corruption, and inefficient allocation of funds to truly come to a
reasonable policy conclusion based on ex-post data.
Still, the main goal of this study is to estimate the effect of HSR on the
reqiured subsidy level according to my model. As noted previously in the
discussion of equation (10), the inclusion of the effects of HSR service on
a given route increase demand by about 23% through the addition of the
hsr variable to the demand function. This increases the vertical intercept to
253 and retains the original slope for a new horizontal intercept of 968,990
passengers, the shift can be seen in figure 5.
Figure 5: German Railway Route- HSR
Published by Digital Commons @ IWU, 2015 33 Undergraduate Economic Review, Vol. 12 [2015], Iss. 1, Art. 4
This also shifts the marginal revenue curve as it is derived from demand
up to a vertical intercept of 253 and again at half the slope of demand, then
the same calcualtions as above are applied to determine fare revenue and
subsidy level.When equating marginal cost with the new marginal revenue
curve, a new passenger level of 431,450 is given. This is then put into
the new demand curve and the original cost curve to yield a fare and cost
per passenger of $112.65 and $125.58 respectively. Then fare revenue and
subsidy is calculated in the same manner as before to yield a fare revenue
of approximately $36,651,700 and a required subsidy of $5,578,650. This
indicates a required annual subsidy that is 10.3% of total annual costs. The
key statistic that I have reached in this section is the difference in required
subsidies between the HSR and non-HSR routes, which is found to be 23.9%,
thus indicating the presence of HSR (through its effect on demand) causes
a 23.9% decrease in required rail subsidy. This is intuitively sound and in
line with my assumed model from Section 3, allowing me to conclude that
the rough model can indeed be applied to the German railway market as a
representation of its basic operation at the route level.
5.2 Model Shortcomings
There are some potential places for improvement and possibilities for
shortcomings in the model however. Most significantly, the magnitudes
of shifts and changes in demand and subsidies are indeed fairly robust,
but are not very specific nor precise. The biggest shortcoming is most
likely in my estimation of average costs, which are understating their actual
levels as discussed above. There are indeed variable costs associated with
https://digitalcommons.iwu.edu/uer/vol12/iss1/4 34 Dorciak: German High Speed Rail
rail travel, but I assumed them away for simplicity of the model, and
with the rationalization that they are relatively small compared to total
infrastructure costs, but nonetheless they are there and would increase
average cost per passenger. Additionally, my level of total route cost is
also understating actual total costs. Realistically, total cost would have to
be measured for each individual route and then regressed to determine a
more fitting estimate for these costs for a given route. This is because I
took a mere average construction cost per mile and then averaging this for
each route, in reality the route construction costs vary wildly in regards
to environmental and urban obstructions that must be negotiated on some
routes, in rare cases being as much as double the average.
Another factor to consider is cost overruns and other expenses that are
not reported in some cases as part of the total construction costs, all of which
combine to indicate that my explicit example understates the total costs of
infrastructure, and in turn average costs per passenger. This indicates that
a more realistic account of costs would then lead to larger required subsidies
to make a route financially viable, combined with my estimated demand. On
the note of demand, in preliminary observation, a logarithmic relationship
would be better suited to my dataset and yields a closer fit, but I proceeded
with the linear regression to maintain model simplicity. Along these lines,
due to the convexity of the logarithmic function and an observation of the
scatter plot of the data points, I would conclude that I slightly overstated
the central range of passenger demand.
All of this combines to indicate what was mentioned previously, in that
most of these studies do not accurately represent the full costs and required
Published by Digital Commons @ IWU, 2015 35 Undergraduate Economic Review, Vol. 12 [2015], Iss. 1, Art. 4
subsidies necessary to support a rail system. However, it is still reassuring to
see that the directions of changes in the model all move as expected. As far
as policy implications, the true decisions lie with policy makers being able
to rationalize the cost to taxpayers against revenues brought in through
fares. The goal of policy makers in this regard is to ensure that demand
and fare revenues are maximized while costs and required subsidies are
minimized, to pass on a minimum of cost to taxpayers and lessen the double
payment made by riders, who essentially pay twice for their fares and in
taxes. This conclusion is not reached in my study, but with a more accurate
representation of costs, the general model could be used to determine this
level and comment on the effectiveness of the implementation of the system
in Germany.
6 Conclusion
Much of the work done to analyze the costs and benefits of HSR to
comment on policy choices presents an overly optimistic view and in many
cases fails to account for the full costs of a rail project. Given a natural
monopoly market structure with fixed costs that exeed revenues for all levels
of demand, it is difficult to comment on the viability of HSR investment
without context as to the financial standing of a country and the ability of
policy makers to justify the cost to taxpayers. Looking at Germany, I worked
to apply a general model of railway subsidization to an ideal representation
of a high speed rail network with readily available data.
In my case, I found that Germany does conform to my model and
https://digitalcommons.iwu.edu/uer/vol12/iss1/4 36 Dorciak: German High Speed Rail
additionally to previous studies such as that by Cuoto and Graham and
exhibits an increase in consumer demand on a given route with the presence
of HSR service. Required subsidies also decrease through a result of my
specified model, but there is much room for improvement upon my basic
framework. Most significantly, an accurate account of all costs combined
with a better specification of costs in the model would yield levels that are
more in line with real world observations of costs that require subsidies of
over 50%.
Assuming that my representation of a given German route is accurate, I
could reasonably reccommend for further high speed development as opposed
to basic railway development and maintenance due to a decrease in subsidies
and increase in fare revenues brought about through increased demand. But
again, before other developments proceed, such as that in California, I would
more conservatively suggest for analysts to look closely at full costs and
any potetnial for cost overruns to not understate cost levels, and also to
make sure an accurate representation of consumer demand is made to not
overstate demand. This will ultimately give policy makers a more accurate
representation of required subsidies and allow them to better justify these
costs to taxpayers.
Published by Digital Commons @ IWU, 2015 37 Undergraduate Economic Review, Vol. 12 [2015], Iss. 1, Art. 4
References
[1] Albalate, Daniel and Germa Bel. ”High Speed Rail: Lessons for Policy Makers from Experiences Abroad.” Public Administration Review, 72 (2012): 336-349. Web.
[2] Behrens, Christiaan, and Eric Pels. ”Intermodal Competition in the LondonParis Passenger Market: High-Speed Rail and Air Transport.” Journal of Urban Economics, 71.3 (2012): 278-88. Web.
[3] Couto, Antonio, and Daniel J. Graham. ”The Impact of High-speed Technology on Railway Demand.” Transportation, 35.1 (2007): 111-28. Web.
[4] ”Deutsche Bahn at a Glance.” Deutsche Bahn. Web. 28 Oct. 2014. http://www.bahn.de/i/view/GBR/en/about/overview/company profile.shtml.
[5] Domencich, Thomas A., and Gerald Kraft. Free Transit. Lexington, MA: Heath Lexington, 1970. Print.
[6] Dunn, James A., and Anthony Perl. ”Policy Networks and Industrial Revitalization: High Speed Rail Initiatives in France and Germany.” Journal of Public Policy, 14.03 (1994): 311. Web.
[7] Eurostat: Density [demo r d2ens]. ”Population density NUTS 3 regions.” Eurostat, (2012). Web. Database code demo r d3dens.
[8] Eurostat: GDP [nama r e2gdp]. ”Gross domestic product at current market prices by NUTS 2 regions.” Eurostat, (2011). Web. Database code - nama r e2gdp.
[9] Eurostat: Railway Transport [tran r rapa]. ”National and international railway passengers transport by loading/unloading NUTS 2 region.” Eurostat, (2010). Web. Database code tran r rapa.
[10] Feigenbaum, Baruch. ”High-Speed Rail in Europeand Asia: Lessons for the United States.”Reason Foundation. Policy Study 418, May 2013. Web.
[11] Garrett, Thomas. ”The Costs and Benefits of Light Rail.” The Central Banker, Fall (2004). Web.
[12] ”Flights - Google Search.” Flights - Google Search. N.p., n.d. Web. 28 Oct. 2014. https://www.google.com/flights/.
https://digitalcommons.iwu.edu/uer/vol12/iss1/4 38 Dorciak: German High Speed Rail
[13] Kuzmyak, J. Richard, and Eric N. Schreffler. Evaluation of Travel Demand Management Measures to Relieve Congestion. U.S. Federal Highway Administration Report No. FHWA-SA-90-005. Springfield, VA: National Information Service. 1990. Print. [14] Levinson, David, Jean Michel Mathieu, David Gillen, and Adib Kanafani. ”The Full Cost of High-speed Rail: An Engineering Approach.” The Annals of Regional Science, 31.2 (1997): 189-215. Web. [15] Mohring, Herbert. ”Optimization and Scale Economies in Urban Bus Transportation.” The American Economic Review, 62.4 (1972): 591-604. Web. [16] ”Germany : DB.” Rail Europe. Rail Europe, n.d. Web. 28 Oct. 2014. http://www.raileurope-world.com/about-us/railways/article/germany-db. [17] Rus, Gins De, and Vicente Inglada. ”Cost-benefit Analysis of the High-speed Train in Spain.” The Annals of Regional Science, 31.2 (1997): 175-88. Web. [18] Savage, Ian. ”A Comment on ’Subsidisation of Urban Public Transport and the Mohring Effect.” Journal of Transport Economics and Policy, 44.3 (2010): 373-80. Web. [19] ”Siemens Receives Order over 15 High-speed Trains from Deutsche Bahn.” Siemens. 17 Dec. 2008. Print. [20] Siemens AG. Communications and Government Affairs. ICE Trains for Deutsche Bahn Approved for Germany. Munich: Siemens, 2013. Print. [21] Siemens, Deutsche Bahn, and the Federal Ministry of Transportation. Deutsche Bahn and Siemens Present New ICE 3 in Berlin. Berlin: Siemens, DB, and BMVI, 2014. Print. [22] Small, Kenneth A. Urban Transportation Economics. Chur: Harwood Academic, 1992. Print. [23] United States. Federation of American Scientists. Congressional Research Service. High Speed Rail in the United States. By David R. Peterman, John Frittelli, and William J. Mallett. Congressional Research Service, 08 Dec. 2009. Web. [24] Utsunomiya, Mariko, and Kenichi Hodota. ”Financial Lessons from Asian Experience in Constructing and Operating High Speed Train Networks.” Transportation, 38.5 (2011): 753-64. Web.
Published by Digital Commons @ IWU, 2015 39